recogna-nlp/EduBench
收藏资源简介:
EduBench是一个巴西葡萄牙语的基准测试,用于评估大型语言模型在教育任务中的性能。该数据集包含3,149个论述性问题,这些问题提取自巴西高竞争性大学入学考试(包括USP、UNICAMP和UNESP),涵盖2015年至2025年共11年的考试。知识领域覆盖人文科学(如历史、地理、哲学、社会学)、精确科学(数学)、自然科学(物理、化学、生物)以及语言(葡萄牙语、文学、英语)。每个问题都包含问题文本、支持文本、完整答案、简短答案、学科分类、难度估计(1-10级)、图像路径和描述等元数据。数据集旨在支持开放式文本生成评估,并提供专家策划的参考答案和丰富的元数据,以促进葡萄牙语NLP研究和教育应用。
EduBench is a comprehensive Portuguese-language benchmark for evaluating large language models on educational tasks. It comprises 3,149 discursive questions extracted from highly competitive Brazilian university entrance examinations (USP, UNICAMP, and UNESP), spanning the period from 2015 to 2025 (11 years of exams). The dataset covers diverse knowledge areas, including Humanities (e.g., History, Geography, Philosophy, Sociology), Exact Sciences (Mathematics), Natural Sciences (Physics, Chemistry, Biology), and Languages (Portuguese, Literature, English). Each question includes the question text, supporting texts, a complete expected answer, a short answer, subject classification, estimated difficulty (1-10 scale), image paths and descriptions, and other metadata. EduBench is designed to assess open-ended, argumentative text generation and provides expert-curated reference answers from official sources, along with rich metadata and automated image descriptions to support text-only evaluation, facilitating research in Portuguese NLP and discursive generation assessment.




